HFIP Global modeling team

Transcription

1 HFIP Global modeling team ESRL Stan Benjamin (+ Mike Fiorino, John Brown, JeffW, Bao ) NCEP/EMC Steve Lord NRL Melinda Peng, Carolyn Reynolds GFDL Shian-Jiann Lin AOML Kevin Yeh Goal: Improve hurricane (tropical cyclone -TC) track (priority 1) and intensity (priority 2) forecasts using improvements to current operational and research global models. HFIP workshop - NHC, Miami, 9-10 Nov 2009

2 Outline Global modeling team goals Stream 1 and Stream 2 Summary of development efforts by ESRL (FIM) EMC (GFS) GFDL (Cubed Sphere Model CSM) NRL (NOGAPS) 2009 TC season HFIP demo for configuration deterministic, ensemble results Key questions raised in 2009 Plans for 2010

3 Statement of the role of global models From HFIP global modeling plan Dec 2008 TC track forecasts for duration of 2 days and longer especially dependent on global models, either directly or through critical lateral boundary conditions for regional models. Thus, HFIP global model forecasts will be the distillation of combined efforts from the Global Modeling Team with those from: Physics Team Data Assimilation Team Ensemble Forecast Team Global Model Team efforts will critically affect Regional Model forecasts for TCs. 3

4 From HFIP global modeling plan Dec 2009 Overview Incorporate improvements to global models from model dynamics, model physics (coordinated with Global Model Physics Team), increased resolution and system diversity and collaborate with other HFIP Teams to achieve HFIP goals Key global modeling components Numerics Horizontal resolution Horizontal grid structure Vertical grid structure Vertical resolution Physical parameterizations Initialization techniques 2010 Global model ensembles 4

5 From HFIP global modeling plan Dec 2009 Overview: Global Modeling Development Team toward improved TC forecasts Stream 1 improve current operational models EMC - development, testing of GFS NRL development, testing of NOGAPS EMC, ESRL - NAEFS readiness for GFS, FIM Stream 2 prepare next-gen global models anticipating substantial HPC augmentation ESRL - development of FIM, demo of high-res FIM, EnKF init of FIM, FIM ensemble, (test new physics) EMC test higher-res GFS on TACC GFDL development/test Cubed Sphere Model NRL development/test advanced NOGAPS 5

6 Outline Global modeling team goals Stream 1 and Stream 2 Summary of development efforts by ESRL (FIM) EMC (GFS) GFDL (Cubed Sphere Model CSM) NRL (NOGAPS) 2009 TC season HFIP demo for configuration deterministic, ensemble results Key questions raised in 2009 Plans for 2010

7 Resolution FIM-2009 model 3 resolutions: 30km, 15km, 10km 64 vertical levels - hybrid theta-sigma Ptop = 0.5 hpa, -top = 2000K ESRL Physics GFS physics suite, many changes to FIM physics interfaces from Nov 2008 May 2009 July upgrades to FIM model for TACC Random number generator for convection cloud top (currently in FIMX) Relaxed vertical regridding much quieter forecasts Distributed output - allows 10km output, faster I/O overall Trisection added to icosahedral grid structure, allowing 10km resolution

8 tighter-than-optimal packing of sigma layers Hybrid sigma-theta pure sigma

9 Formation of Typhoon Choi-Wan 10km FIM ESRL

10 Formation of Typhoon Choi-Wan 10km FIM ESRL

11 Formation of Typhoon Choi-Wan 10km FIM ESRL

12 Formation of Typhoon Choi-Wan 10km FIM ESRL

13 ESRL FIM Changes in July-Aug 2009 Version updates for GSD FIM: 8/26/ z - FIM revision Includes modified hybgen (vertical filter, minimum thickness near top, increased relaxation in coordinate movement toward target values). Setting of ptop to 50 Pa (instead of 10 Pa) and min thickness near top = 50 Pa. TACC FIM updated to r762 at same time. 7/26/ z - FIM revision Includes xkt2 random number generator cloud top in convective parameterization, also fixes errors to properly set the initial liquid soil moisture (was being set as total soil moisture including frozen). 7/22/ z - Revision to reference potential temperature table to spread out vertical levels near top, revision 661 3/20/ z - FIM revision Correction of initialization and evolution of liquid soil moisture.

14 FY10 Implementation Plan NCEP GSI/GFS Fall Bundle December Ingest new data types GSI Code changes GFS code restructuring Consolidation of GDAS & GFS Posts Benefits Better tropical cyclone definition Small incremental improvement in forecast skill GFS Shallow Convection March Shallow convection, Deep Convection, PBL Benefits Significant reduction in gridpoint storms Small incremental improvement in forecast skill GFS Resolution Increase May T574L64 Benefits Overall improvement in forecast skill 14

15 GSI Changes December 2009 NCEP Adding new observation data sources. Tropical storm pseudo sea-level pressure obs NOAA19 hirs/4,amsu-a, & MHS brightness temp obs NOAA18 sbuv/2. Monitor N19 GOME, and OMI ozone (no assimilation) RARS (currently only EARS) 1B data EUMETSAT-9 atm motion vectors Implementing improved techniques in GSI analysis. Use uniform thinning mesh for brightness temp data Improvements to assimilation of GPS RO data (QC, retune ob errors, improved forward operator ) Add dry mass pressure constraint Merge GMAO & EMC codes for 4d-var capability Update background error covariance Proper use of different spectral truncation between background and analysis 15

16 Expected Benefits NCEP Data Assimilation Improve tropical storm track & intensity forecasts Slight incremental improvement in overall forecast accuracy Model Facilitate testing and implementation of future upgrades Reduce code maintenance Improvements expected to be small and primarily due to the assimilation changes Details on assimilation effects on GFS TC forecasts from Steve Lord (assim) and Carolyn R. (ensemble) 16

17 Experiments for HFIP NCEP High resolution global ensembles (NCEP/GEFS) T574L64 (~23km horizontal resolution) Initial analysis GSI T382L64 analysis ETR (ensemble transform with rescaling) Every 24 hours Cycling at T382L64 resolution NCEP/CCS Upgrade to T574L64 Integrations At Texas Advanced Computing Center (TACC) Use GFS model at T574L64 resolution 5 members (include control) Out to 168 hours Experiments Once per day for period of Sept. 1 st 20 th 2009 Output Tracks for each members, ensemble mean

18 NRL Global Model Improvements NRL Data Assimilation and Global Model Personnel NRL Global Forecasting System Major Upgrade September From 30 levels (left) 3DVAR to 42 Levels and 4DVAR

19 NRL NRL NOGAPS Recent Upgrades: Impact on TC track New multi-level sea-ice temperature prediction, Change in lower level cumulus cloud prediction (fixed low cloud bias) Correct optical thickness calculation, which over-estimated mixed cloud thicknesses Increased lower perturbation temperature in the cumulus parameterization Results Statistically significant improvement in SH AC Statistically significant improvement in TC tracks at 96 h Small but significant Arctic AC improvement in the winter season Reduced lower tropical temperature error No increase in run-time Ridout, Hogan, Peng

20 The GFDL High-Resolution Atmosphere Model is developed for km resolution, sharing most of the codes with the GFDL AM2/AM3 for IPCC, except the following major modifications GFDL Non-hydrostatic Cubed-sphere Finite-Volume dynamical core. 6-category single-moment bulk cloud microphysics with computational efficiency significantly improved with time implicit treatment of microphysics processes and vertically Lagrangian terminal fall of all condensates (rain, snow, ice, and graupel). Subgrid cloud formation is accounted for by using both vertical & horizontal subgrid distributions (with a PDF approach). The deep convective parameterization scheme is replaced by an essentially non-precipitating shallow convection scheme (Zhao et al. 2009) Surface fluxes modified for high-wind situation over ocean (Moon et al. 2007)

21 Gnomonic Cubed Sphere grid GFDL Defined by intersects of great circles with equaldistance along 12 edges Maximum local grid aspect ratio ~ Maximum global grid aspect ratio ~ Resolutions: 1) TC climatology: C180, x ~ 50 km 2) HFIP 2009: C360, x ~ 25 km 3) HFIP 2010: C720, x ~ 12.5 km 4) Cloud-resolving C2000, x ~ 4.5 km Can also be used as a regional model Hurricane in a doubly periodic box

22 Outline Global modeling team goals Stream 1 and Stream 2 Summary of development efforts by ESRL (FIM) EMC (GFS) GFDL (Cubed Sphere Model CSM) NRL (NOGAPS) 2009 TC season HFIP demo for configuration deterministic, ensemble results Key questions raised in 2009 Plans for 2010

23 NOAA global model experiments - TACC-Ranger Austin, TX 62,976 processor machine using Linux NOAA allocated 15 million hours from 1 July through 31 Dec 2009 What actually happened 11 million hours used 7/15/ /2/ % - EnKF data assimilation 6h update cycle -9h forecasts 20% - 10km FIM once daily 20% - FIM30 20-member ensemble once daily 20% - FIM15 2 runs EnKF, GSI initial conditions Limited runs of GFS ensemble T574/64L 5-member ensemble 4 TCs 4 Mh reserved for 2008 EnKF Work underway Queue now very slow compared to Real-time HFIP demo period 23

24 FIM, GFS, 500 hpa anom corr N. Hemisphere 120h FIM ~ GFS, will contribute strongly to GEFS FIM better GFS better

25 120h 500hPa height Anom Corr Coef (ACC) Aug09 N.Hem FIM GFS GSI init conds EnKF init conds S.Hem FIM slightly better than GFS EnKF > GSI IC

26 72h 850hPa Wind RMS vector error (vs. analyses) smaller is better N.Hem FIM GFS GSI init conds EnKF init conds Tropics FIM much better better than GFS than in NH GFS ( and in SH), NH/SH, not tropics EnKF IC better

27 72h 250hPa Wind RMS vector error (vs. analyses) smaller is better N.Hem FIM GFS GSI init conds EnKF init conds S.Hem FIM better than GFS in NH/SH, EnKF IC adds further accuracy

28 72h 250hPa Wind RMS vector error (vs. analyses) smaller is better Tropics FIM GFS GSI init conds EnKF init conds FIM better than GFS, EnKF IC adds further accuracy

29 2009 Atlantic hurricanes (near real-time forecasts) GFDL ( forecasts by Jeff Gall )

30 2009 near real-time forecasts GFDL Track errors Intensity errors (forecasts by J. Gall, analysis by T. Marchuk)

31 TC intensity forecast error all basins, FIM-15km-det vs. FIM-15-ens-mean - GFDL CSM added FIM-15km-EnKF similar in intensity forecast to GFDL CSM FIM-15km-det-EnKF much better than FIM-30km-mean for intensity ESRL 31

32 TC intensity forecast error all basins, versions of FIM-15km: GSI init conds, EnKF init conds - FIM 10km vs. 15km vs. 30km ESRL Clear improvement for FIM w/ EnKF IC 10km best intensity FIM better intensity than GFS even at 30km 32

33 TC track forecast error all basins, FIM-15km-det vs. FIM-15-ens-mean - GFDL CSM added FIM-15km-det-EnKF slightly better than FIM-30km-mean for track for 48, 72, 96h, slightly worse for 120h FIM track forecasts better than CSM-C360 at all forecast times GFS has lowest track error for this limited number of cases ESRL 33

34 TC track forecast error all basins, versions of FIM-15km: - GSI init conds, EnKF init conds ESRL Track fcst error n. miles All basins-2009 Clear improvement for EnKF IC FIM-EnKF better than GFS 72-96h, FIM-GSI worse than GFS 34

35 Outline Global modeling team goals Stream 1 and Stream 2 Summary of development efforts by ESRL (FIM) EMC (GFS) GFDL (Cubed Sphere Model CSM) NRL (NOGAPS) 2009 TC season HFIP demo for configuration deterministic, ensemble results Key questions raised in 2009 Plans for 2010

36 Current & plans: GFDL C360 (~25 km) resolution for seasonal hurricane predictions C720 (~12 km) for near future (2010?) 5-10 day forecasts Coupled GFDL ocean and Wave Watch III model (to improve surface momentum & heat fluxes) Improve the 4D vortex initialization (and reducing the initial position & intensity errors!) Global cloud-resolving (~4.5 km with minimal convective parameterization) forecast experiments to be performed at DOE s Argonne National Laboratory. Platform: IBM Blue Gene (P/Q) scaling to 200,000 cores (or higher) required for real-time forecast to be meaningful.

37 NRL NOGAPS Future Plans: FY10 Semi-Lagrangian/Semi-implicit scheme (higher resolution) Improved Convective Momentum Transport Above two topics in global physics summary New Fu-Liou radiation scheme Improved cloud prediction ESMF superstructure framework Upgrade of the EFS Convective and mid atmosphere GWD

38 FIM next year Implementation at NCEP in 2010 as experimental component for global ensemble forecast system using ESMF (Henderson, HFIP engineer, EMC) Improved numerics continuity eqn, vertical coordinate Further testing of WRF physics options (Grell/Devenyi cumulus, Lin microphysics), WRF/chem options: (Grell, Bao). Additional work on OFIM - icosahedral matched HYCOMisopycnal ocean component Continue testing with GOCART aerosols, real-time fire data with global emissions inventories ESRL Initialize with improved initial conditions EnKF with improved TC balance, EnKF with both GFS and FIM Continued development of 5km Nonhydrostatic Icosahedral Model (NIM) (Lee, MacDonald, others ESRL)

39 FIM Work underway: Develop mirror FIM-HYCOM Atmosphere-Ocean on same icosahedral grid - OFIM CFIM HYCOM ESRL Advantages: avoid atmos/ocean coupling problems (1-1 pt correspondence) Inherits: icos numerical advantages FIM model framework (modules, MPI) augmented w/ HYCOM ocean prognostic eqns ESMF/NUOPC compliance 70 N 75 S

40 Initial FIM tests w/ WRF physics resolved Grell-Devenyi convection Lin microphysics ESRL 60km- 12hr total precipitation comparison Grell-Devenyi conv schcheme GFS Physics w/ Pan- Grell conv scheme

41 FIM- with GOCART parameterization (18 aerosols) + GFS physics ESRL Dust and Sea-salt, 5-day forecast G7-60km resolution tests here Saharan Dust mid-level Total PM2.5 (dust and sea-salt, some sulfate) at surface

42 Global Model Team HFIP 2009 achievements High-resolution real-time demonstrations GFS-T572 (23km), FIM-10km/15km, CSM-C360/720 (25km/12km) High-res ensemble FIM-30km / 20 member Improved numerics in FIM vertical regridding Development toward ESRL NRL NOGAPS-SL testing New physics / data-assimilation tested with experimental high-res global models global scale Higher resolution more accurate intensity forecasts but not necessarily more accurate track (yet) Demonstration of improved track forecasts from EnKF assimilation together with high-res global model (FIM, GFS) TC demographics not revealed by current track/intensity verification, but major emphasis in 2009 HFIP global model (and physics) teams Exposure to year-round testing reveals unanticipated results FIM polar-night stratopause CFD instability (solved) FIM-EnKF exposure of surface pressure noise issue, now leading to improved FIM continuity equation solution

43 Primary global modeling issues Performance for TC track, intensity, demographics Maintain at least equal performance in extratropics and polar regions while improving tropical performance Numerics - understand effects on TC track, intensity, demographics Horizontal and vertical coordinates Diffusion Resolution Not yet fully addressed in 2009 topography and landsurface not yet introduced at sub-45km resolution in FIM Coupled atmos-ocean model non-linearity Interdependencies between all issues above with physics and data assimilation Interaction w/ Physics Team and Data Assimilation Team (much stronger in 2009 than in 2008)